BARC

Overview

This repository accompanies the paper:

When T-Depth Misleads: Predicting Fault-Tolerant Quantum Execution Slowdown under Magic-State Delivery Constraints

This paper has been submitted to QCE 2026.

The artifact studies how circuit dependency structure interacts with a fixed delivery rate C and a fixed buffer capacity B, and packages the paper-facing figures and tables used by the manuscript.

Problem

Fault-tolerant quantum compilers often optimize for static T-depth, but bounded magic-state delivery changes what actually runs fast. A schedule can look shallow on paper and still execute poorly if it creates bursts of T-gate demand that exceed delivery capacity and force protected waiting.

Contribution

This artifact evaluates that mismatch using one structural indicator, slack_ratio, and one schedule-level indicator, delta_max. It shows when T-depth becomes misleading, validates a fixed-schedule lower bound on executable makespan, and packages the paper-facing figures and tables behind those claims.

Key Figure

The figure below illustrates the core problem. The blue curve is cumulative T-state demand from a compiled schedule, while the dashed line is the cumulative supply envelope under delivery rate C and buffer B. When demand rises above supply, backlog accumulates. That backlog delays execution even if the original static schedule is short. The peak gap, delta_max, is the main schedule-level indicator used in the paper.

Paper figure files: delta_max_illustration.png, delta_max_illustration.pdf

The main outputs of the artifact are:

• a structural metric based on T-node slack
• a system-level metric delta_max
• a fixed-schedule lower bound on executable makespan

The repository is organized around four layers:

• model: dependency DAGs and demand traces
• scheduling: static schedule construction
• simulation: bounded-delivery execution
• analysis: predictive evaluation and figure generation

Key Concepts

• slack_ratio: fraction of T nodes with positive slack in the dependency DAG
• mean_t_slack: mean slack over T nodes
• delta_max: peak prefix deficit between cumulative T demand and cumulative delivery capacity
• slowdown_ratio: T_exe / T_static

For a fixed valid schedule with static depth T_static, delivery capacity C, buffer B, and prefix deficit delta_max, the artifact evaluates the lower bound

T_static + ceil(max(0, delta_max - B) / C)

against the simulated executed makespan T_exe.

Reproducibility

Install dependencies:

Dependency file: barc_stage1/requirements.txt

python -m pip install -r barc_stage1/requirements.txt

Run the full paper pipeline from the artifact root:

cd barc_stage1
python scripts/run_qce_paper.py

Run the test suite:

cd barc_stage1
python -m unittest discover -s tests -p 'test_*.py'

Prepare a clean manuscript bundle containing the final PDF figures and key tables referenced by the paper:

Bundle script: barc_stage1/scripts/prepare_submission_bundle.py

python barc_stage1/scripts/prepare_submission_bundle.py

Optional appendix robustness assets can be regenerated with:

cd barc_stage1
python scripts/run_stochastic_supply_sensitivity.py
python scripts/run_routing_proxy_sensitivity.py
python scripts/build_appendix_robustness_figure.py

The main pipeline generates:

• paper-facing tables in outputs/tables/
• paper-facing figures in outputs/figures/
• secondary tables and figures in outputs/appendix/

Submission Assets

The manuscript-facing files are split into two groups:

• final PDF figures for LaTeX: barc_stage1/outputs/figures/ and barc_stage1/outputs/figures/final_paper/
• quantitative tables behind the reported claims, including appendix support tables: barc_stage1/outputs/tables/

Running python barc_stage1/scripts/prepare_submission_bundle.py creates submission_bundle/ at the repository root with:

• figures/ laid out to match the LaTeX \includegraphics{figures/...} paths
• tables/ containing the main CSV/TXT assets needed to audit manuscript claims
• MANIFEST.md summarizing every copied file

Manuscript Figure Map

These are the figure PDFs directly referenced by the current paper draft.

• figures/delta_max_illustration.pdf
• figures/final_paper/predictor_comparison_final.pdf
• figures/final_paper/incremental_predictive_gain_final.pdf
• figures/final_paper/structure_to_execution_chain_empirical_final.pdf
• figures/final_paper/delta_max_vs_slowdown_final.pdf
• figures/final_paper/lower_bound_vs_actual_final.pdf
• figures/final_paper/qft_vs_other_real_traces_final.pdf
• figures/final_paper/real_trace_scaling_final.pdf
• figures/final_paper/qft_approximation_reduced_grid_final.pdf
• figures/final_paper/robustness_stochastic_routing_summary_final.pdf
• figures/final_paper/lower_bound_gap_cases_final.pdf

Manuscript Table Map

These files are the primary quantitative sources for the current draft.

• family_summary.csv: family-level slowdown and Delta_max summaries
• inversion_summary.csv: T-depth inversion rates
• predictive_classification_summary.csv: stall and inversion AUC summaries
• predictive_regression_summary.csv: slowdown correlation summaries
• predictive_multivariate_regression.csv: representative multivariate slowdown fits
• incremental_predictive_models.csv: incremental gain values used in the predictor discussion
• bootstrap_slack_vs_tdepth.csv: paired bootstrap confidence intervals
• causal_chain_correlations.csv: structure-to-system-to-execution correlation chain
• lower_bound_validation.csv: 4,904 finite instances used in lower-bound validation
• lower_bound_gap_cases.csv: representative finite positive-gap cases
• qft_real_trace_summary.csv: exact-QFT real-trace summary values
• real_trace_scaling_summary.csv: adder/multiplier scaling values
• qft_approximation_reduced_grid_summary.csv: exact vs approximate QFT reduced-grid comparison
• stochastic_supply_ranking_summary.csv: appendix-level stochastic supply sensitivity summary
• routing_proxy_ranking_summary.csv: appendix-level routing proxy sensitivity summary
• delivery_aware_pass_probe_round2_summary.csv: appendix-level preliminary compiler-probe summary

Main Results

The primary figures for the paper are:

• slack_vs_slowdown.png
• delta_max_vs_slowdown.png
• slack_vs_stall.png
• structure_to_execution_chain.png
• lower_bound_vs_actual.png

The primary tables are:

• stage1_grid_scan.csv
• family_summary.csv
• predictive_classification_summary.csv
• predictive_regression_summary.csv
• predictive_multivariate_regression.csv
• causal_chain_summary.csv
• causal_chain_correlations.csv
• lower_bound_validation.csv
• predictive_analysis_summary.txt
• paper1_reframing_summary.txt

The real-trace grounding remains available under src/real_trace/ and outputs/real_trace/. In the paper narrative, these traces serve as grounding examples rather than as the main source of method comparison.

Scope and Limitations

The artifact uses:

• deterministic delivery capacity
• deterministic buffer capacity
• dependency-aware synthetic DAG families
• fixed schedule policies

The main deterministic paper pipeline does not include:

• routing or layout effects
• stochastic delivery
• noise, decoding, or full physical simulation
• optimal scheduling over all valid schedules

Separate appendix robustness scripts provide first-order sensitivity checks for stochastic supply and route-induced effective-capacity proxies. The lower bound remains a fixed-schedule deterministic result. It does not characterize optimal scheduling across all valid schedules and it does not minimize delta_max over the full schedule space.

The repository also contains a preliminary quota-respecting scheduling probe. This probe is included only as appendix support for the current manuscript and should not be interpreted as a full compiler evaluation.

Repository Layout

The source tree keeps the paper pipeline compact:

• barc_stage1/src/dag.py, barc_stage1/src/trace.py: model layer
• barc_stage1/src/schedule.py: scheduling layer
• barc_stage1/src/simulator.py: simulation layer
• barc_stage1/src/metrics.py, barc_stage1/src/predictive_analysis.py: analysis layer
• barc_stage1/src/plots.py: figure generation
• barc_stage1/src/stochastic_supply.py: stochastic supply sensitivity utilities
• barc_stage1/src/robustness_workloads.py: representative workload selection for appendix robustness studies
• barc_stage1/src/real_trace/: real-trace grounding utilities
• barc_stage1/src/real_trace/circuit_to_dag.py: convert real Qiskit circuits into the internal DAG representation used by scheduling probes
• barc_stage1/scripts/run_qce_paper.py: single entry point
• barc_stage1/scripts/run_stochastic_supply_sensitivity.py: stochastic supply robustness study
• barc_stage1/scripts/run_routing_proxy_sensitivity.py: route-induced effective-capacity proxy study
• barc_stage1/scripts/build_appendix_robustness_figure.py: combine robustness summaries into a manuscript-facing PDF figure
• barc_stage1/scripts/run_delivery_aware_pass_probe_round2.py: appendix-level preliminary compiler-probe summary generation
• barc_stage1/tests/: slack validation tests

Appendix Outputs

Non-primary outputs are written to outputs/appendix/. This includes:

• policy comparison artifacts
• QTV tradeoff artifacts
• family-level gain boxplots

These files are retained for completeness, but they are not part of the main paper narrative.